Dairyland powers up for a generative AI edge

Under the leadership of CIO Nate Melby, the Midwestern utility has been an early mover in putting machine learning and LLMs to work to deliver safe, reliable, and affordable electricity to its members.

A Midwestern utility cooperative might not be the first place you’d look for leading-edge implementations of emerging technologies, but thanks to the leadership of CIO Nate Melby, Dairyland Power Cooperative has become an unlikely pioneer in generative AI, churning out large language models (LLMs) that not only automate document summarization but also help manage power grids during storms.

Previously head of cybersecurity at Ingersoll-Rand, Melby started developing neural networks and machine learning models more than a decade ago. He brought that experience with him to Dairyland in 2016 when he was appointed as the cooperative’s first CIO to oversee 24 power grids in Wisconsin, Iowa, Illinois, and Minnesota.

Melby pushed machine learning models into production very early at Dairyland, improving the cooperative’s weather forecasting capabilities and creating load management applications that “bent the curve” to best manage the company’s power load on peak days, the CIO says.

Now, thanks to the cooperative’s tight partnership with Microsoft systems integrator Stoneridge Software, as well as Melby’s extensive technology experience, Dairyland — which was formed during the New Deal in the 1930s — has been able to experiment with and put into production some of the earliest Microsoft Azure-based LLMs, Melby says.

“We were positioned correctly. We just came out of the gates fast, and we just kept solving problems,” the CIO says, noting that his team was experimenting with Azure LLMs before they were on the market. “We were trying to take advantage of the technology and make the right moves. We did not realize we were that far ahead.”

Laying the foundation

The La Crosse, Wisc.-based power utility is owned by 24 member systems, serving 700,000 people in rural areas of the upper Midwest. When Melby took on the role of CIO, he saw an organization that was seriously in need of digital transformation.

Dairyland’s early implementation of AI was a byproduct of its work in Microsoft’s FastTrack program, which saw Dairyland shifting to Dynamics 365 ERP and the Azure cloud. Melby believes his deep cybersecurity background in the enterprise and grasp of the imperative for data privacy in AI enabled him to earn the trust of Microsoft and work in the inner circle and at the leading edge.

“We put in the request in the first hour of the first day we knew we could and knew it was a bit crazy because we were in as soon as some of other huge companies with tons of resources,” says Melby, who worked closely with Stoneridge Software to develop its digital platform.

Beginning in 2021, the Minneapolis-based Microsoft partner helped Dairyland migrate from several custom legacy applications to a commercial implementation of Dynamics 365 and an Azure data lake, which set the stage for the power company’s early foray into AI, according to the systems integrator.

“Dairyland has been on this journey and done this digital transformation to the whole Azure Microsoft platform to elevate them in the future to do the AI work,” says Dave Ruelle, a partner at Stoneridge Software, which is working with a variety of clients to implement Microsoft Copilot. Melby’s AI development for Dairyland, however, was all performed in-house, Ruelle notes. “I would say they’re at the forefront of it.”

Melby jumped on AI quickly because the cooperative is under pressure to provide safe and reliable electricity to members at the lowest rate possible. Drawing on his experience, deep technical insight, and doctorate in information systems, Melby had several practical ideas right away about how to apply machine learning models and generative AI to serve that mission and to produce other business efficiencies.

“From the moment that it was available, we just took it and ran with it,” the CIO says. “I was literally just waiting for commercial availability [of LLMs] but [services] like Azure Machine Learning made it so you could easily apply it to your data. It was cloud-based and something you could commercially license.”

Jumpstarting innovation

Initially, Dairyland jumped at a more generalist use of GPT-3 and GPT-4.5 LLMs. The team developed an AI-based writing assistant with analysis features that was initially used in communications and marketing to develop press releases and social media posts, Melby says.

After kicking the tires on other Azure LLMs, software developers on Melby’s team saw development time for certain workflows dwindle from two weeks to 15 minutes. One SCADA engineer, for example, started using generative AI “to create synthetic data for testing our SCADA systems,” Melby says.

“As a team, we knew one good way for us to start surfing the wave of this AI tsunami was with a general assistant, and when we made it available internally, there was huge interest and use,” says Melby, noting that, as his team of roughly 70 IT pros experimented with the public API, the “use cases started pouring in.”

Vladimir Tsoy, business applications supervisor at Dairyland, says Melby’s strategy and “progressive stance on technology” has been a “game changer” for the power company.

“The energy utilities sector isn’t known for being the quickest to embrace emerging technologies, but Nate’s leadership has brought us to the forefront of innovation,” Tsoy says, adding that Melby’s leadership has been instrumental in getting its AI initiatives up and running. “He was aware of some of the projects I had been working on for some time, and with his support, we managed to roll out some of these [AI applications] into production pretty quickly.”

Dairyland leans heavily on Microsoft but also maintains some applications on a private cloud and data centers for reliability, as part of the critical infrastructure, the CIO says.

Melby was nominated by the WisconsinCIO Advisory Board as a finalist for CIO of the year based on his leadership, effective management, and business value created through innovation, says Christa Ogilvy, executive director of ChicagoCIO and WisconsinCIO. He is “part of the brain trust” that hosted an AI strategy discussion with more than 30 CIOs in February and is one of 40 CIOs in Wisconsin who works collaboratively with peers to solve problems.

Looking ahead, Melby sees AI making a big impact on the industry’s biggest unsolved problems.

“In the electric utilities, it’s improved reliability for our generation assets, increased efficiency in our transmission systems, and is helping us be more sustainable by managing the energy transition with more renewables coming onto the grid,” Melby says, adding that “finding ways for us to make safety improvements would also be a huge benefit.”

To date, Dairyland is strategizing on ways to leverage retrieval-augmented generation (RAG), vector databases, and multimodality to create efficiency, the CIO says.

“For example, can we learn about our work environment using AI to analyze our own team’s observations of hazards and actions in the field, and use this to help us focus on the right things to improve our safety performance?” Melby conjectures. “These are the kinds of things we’re looking at, and it starts with identifying the problems that we want to solve.”

SUBSCRIBE TO OUR NEWSLETTER